[LU-9377] obdfilter-survey.sh test_3a case=network is not functional on setup with N oss servers

• Status: Open
• Project: Lustre
• Type: Bug
• Priority: Minor
• Reporter: Arshad Hussain
• Assignee: Emoly Liu
• Resolution: Unresolved
• Severity: 3

Description

obdfilter-survey when case is 'network' is failing because multiple IP/hostname are supplied as input.

stdout.log
+ NETTYPE=tcp thrlo=8 nobjhi=1 thrhi=16 size=1024 case=network rslt_loc=/tmp targets="192.168.109.6 192.168.109.5" /usr/bin/obdfilter-survey
Permission denied, please try again.

Permission denied, please try again.

Comments

[Gerrit Updater]

Arshad Hussain ([e-mail]) uploaded a new patch: https://review.whamcloud.com/26768
Subject: LU-9377 tests: Fix obdfilter-survey/3a to take single IP
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 001d6baaae9695eda731a0fa86ccb368b3e6e1b2

[Peter Jones]

Emoly

Could you please review this patch?

Thanks

Peter

[Emoly Liu]

Actually, the "case='network'" does only support single IP/hostname.
If you want to support multiple IP/hostnames in test, you should improve some functions in files lustre-iokit/obdfilter-survey/{obdfilter-survey, iokit-libecho}, but not obdfilter-survey.sh.

BTW, I just tested your patch with two OSS nodes, it showed

[root@trevis-73vm7 tests]# mgs_HOST=trevis-73vm3 MGSDEV=/tmp/lustre-mgs MDSCOUNT=2 mds1_HOST=trevis-73vm3 mds2_HOST=trevis-73vm4 ost1_HOST=trevis-73vm5 ost2_HOST=trevis-73vm6 OSTCOUNT=2 PDSH="pdsh -S -Rrsh -w" ONLY=3 sh obdfilter-survey.sh
Logging to local directory: /tmp/test_logs/1493005947
excepting tests: 1b 2b
trevis-73vm7.trevis.hpdd.intel.com: Checking config lustre mounted on /mnt/lustre
Checking servers environments
Checking clients trevis-73vm7.trevis.hpdd.intel.com environments
Using TIMEOUT=20
disable quota as required
osd-ldiskfs.track_declares_assert=1
osd-ldiskfs.track_declares_assert=1
osd-ldiskfs.track_declares_assert=1
osd-ldiskfs.track_declares_assert=1
min kbytesavail: 309264 using size=251 MBytes per obd instance


== obdfilter-survey test 3a: Network survey ========================================================== 03:52:33 (1493005953)
Stopping clients: trevis-73vm7.trevis.hpdd.intel.com /mnt/lustre (opts:)
Stopping client trevis-73vm7.trevis.hpdd.intel.com /mnt/lustre opts:
Stopping clients: trevis-73vm7.trevis.hpdd.intel.com /mnt/lustre2 (opts:)
Stopping /mnt/mds1 (opts:-f) on trevis-73vm3
Stopping /mnt/mds2 (opts:-f) on trevis-73vm4
Stopping /mnt/ost1 (opts:-f) on trevis-73vm5
Stopping /mnt/ost2 (opts:-f) on trevis-73vm6
Stopping /mnt/mgs (opts:) on trevis-73vm3
modules unloaded.
+ NETTYPE=tcp thrlo=2 nobjhi=1 thrhi=4 size=251 case=network rslt_loc=/tmp targets="" /bin/obdfilter-survey

targets="", this is not correct.

[Colin Faber [X] (Inactive)]

Hi, is this still active?

[Arshad Hussain]

Hi Colin,

>Hi, is this still active?

Could not follow up on Emoly's review comment. This patch was not actively worked on.

The motivation of the patch was to accept single IP when parameter "case=network" is given and not fail. The failure seen was inability to correctly check "obdecho" module on server.

Currently, when multiple IP is provided, we "mess" up the parameters passed and end up having

192.168.50.95@192.168.50.95 root 'lsmod | grep obdecho'

instead of more correctly (I think so)

192.168.50.95 root 'lsmod | grep obdecho'

Based on Emoly's review. (this is where we stand today)
1. Current patch does not work

2. "case='network'" currently only support single IP/hostname. Therefore, easy fix would be to either not permit multiple ip when case=network. So it does not fail. (This is what this patch intended to do). Further, Update doc pdf.

3. Major improvement to make it accept multiple IP (improve iokit-libecho/setup_srv_obd), which this patch does not target.